Ensemble Learning Prediction Model for Intraoperative Hemodynamic Instability in Patients with Pheochromocytoma

Yingshu Liu^1,2Chao Liu³Shen Li²Liang Zhao⁴Rui Lin⁴Kang Chen¹Li Zang¹Weijun Gu¹Yiming Mu¹Zhao Hui Lyu¹Zhengnan Gao¹Jingtao Dou^1*

• ¹1st Medical Center of Chinese PLA General Hospital, Beijing, China
• ²Central Hospital of Dalian University of Technology, Dalian, China
• ³Yidu Cloud Technology Co Ltd, Beijing, China
• ⁴Dalian University of Technology, Dalian, China

Background Accurately predicting intraoperative hemodynamic instability (HI) in patients with pheochromocytoma is essential for improving prognosis; however, clinically applicable large-sample, high-precision predictive models remain limited. This study develops and validates an ensemble learning (EL) model to predict HI risk. Methods This cohort study included a derivation cohort (n = 353) and an external validation cohort (n = 51), from January 2011 to February 2023. General clinical and intraoperative hemodynamic data were collected. Ensemble feature selection was used to identify key predictors. 5-fold cross-validation was repeated 1000 times to develop the EL model. Shapley Additive Explanations was used to analyze feature contributions, and the model was implemented as a web calculator. The primary outcome was the occurrence of intraoperative HI, evaluated by area under the curve (AUC), sensitivity, specificity, and calibration. Results Of 45 variables, tumor size, preoperative systolic blood pressure, age, fasting plasma glucose, and body mass index were top predictors. The developed EL model achieved AUC, sensitivity, and specificity values of 0.886, 0.776, and 0.836 and 0.744, 0.733, and 0.667 in Training set and external validations, respectively. Higher SBP (≥ 125 mmHg), larger tumor size (≥ 60 mm), older age (≥ 55 years), higher FPG (≥ 6 mmol/L), and BMI <22 or >30 kg/m² increased HI risk. The model demonstrated strong calibration and is accessible at http://60.205.91.235/. Conclusions